Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Cognitive Explainer for Fetal ultrasound images classifier Based on Medical Concepts

Yingni Wanga, Yunxiao Liua, Licong Dongc, Xuzhou Wua, Huabin Zhangb, Qiongyu Yed, Desheng Sunc, Xiaobo Zhoue, Kehong Yuan

TL;DR

This work proposes an interpretable framework based on key medical concepts, which provides explanations from the perspective of clinicians' Cognition, and utilizes a concept-based graph convolutional neural(GCN) network to construct the relationships between key medical concepts.

Abstract

Fetal standard scan plane detection during 2-D mid-pregnancy examinations is a highly complex task, which requires extensive medical knowledge and years of training. Although deep neural networks (DNN) can assist inexperienced operators in these tasks, their lack of transparency and interpretability limit their application. Despite some researchers have been committed to visualizing the decision process of DNN, most of them only focus on the pixel-level features and do not take into account the medical prior knowledge. In this work, we propose an interpretable framework based on key medical concepts, which provides explanations from the perspective of clinicians' cognition. Moreover, we utilize a concept-based graph convolutional neural(GCN) network to construct the relationships between key medical concepts. Extensive experimental analysis on a private dataset has shown that the proposed method provides easy-to-understand insights about reasoning results for clinicians.

A Cognitive Explainer for Fetal ultrasound images classifier Based on Medical Concepts

TL;DR

This work proposes an interpretable framework based on key medical concepts, which provides explanations from the perspective of clinicians' Cognition, and utilizes a concept-based graph convolutional neural(GCN) network to construct the relationships between key medical concepts.

Abstract

Fetal standard scan plane detection during 2-D mid-pregnancy examinations is a highly complex task, which requires extensive medical knowledge and years of training. Although deep neural networks (DNN) can assist inexperienced operators in these tasks, their lack of transparency and interpretability limit their application. Despite some researchers have been committed to visualizing the decision process of DNN, most of them only focus on the pixel-level features and do not take into account the medical prior knowledge. In this work, we propose an interpretable framework based on key medical concepts, which provides explanations from the perspective of clinicians' cognition. Moreover, we utilize a concept-based graph convolutional neural(GCN) network to construct the relationships between key medical concepts. Extensive experimental analysis on a private dataset has shown that the proposed method provides easy-to-understand insights about reasoning results for clinicians.
Paper Structure (22 sections, 6 equations, 5 figures, 5 tables)

This paper contains 22 sections, 6 equations, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Medical concepts for fetal standard scan plane diagnosis
  4. Machine-aided diagnosis of fetal US planes
  5. Interpretability methods in medical application
  6. Graph Neural Networks
  7. Method
  8. Medical concepts identification with prior knowledge
  9. Concept graph construction
  10. Concept graph Learning
  11. Post-hoc graph explainer
  12. Datasets and Implementation details
  13. Datasets
  14. Implementation details
  15. Evaluation metrics
  16. ...and 7 more sections

Figures (5)

  • Figure 1: Fetal abdominal, femoral, and thalamic anatomy (the top row) and typical standard scan planes(the bottom row).
  • Figure 2: Overview of the interpretability framework. We first extract class-specific medical concepts approved by sonographers with prior medical knowledge. The medical concepts are transformed into graph-structured data, and use GCN to learn the contribution of nodes (medical concepts) and edges (relationships between concepts) to decision-making and to explain the decision-making process of the network.
  • Figure 3: Medical concept reasoning explanation of Fetal US Standard Plane for MobilenetV2 in B dataset. The rows represent the FSP types, i.e. FASP, FFSP, and FTSP, and the columns represent the graph interpretability methods, i.e. Graph SA, Graph IG and Graph Grad-CAM. Concept importance ranges from blue (the least important) to red (the most important).
  • Figure 4: Comparison of explainability methods on Fetal Standard Planes based on MobilenetV2.
  • Figure 5: Comparison of explainability methods on Fetal Standard Planes based on ResNet34.